1 monitoring spatial and temporal changes of ...

1

MONITORING SPATIAL AND TEMPORAL CHANGES OF AGRICULTURAL LANDS IN THE NILE DELTA AND THEIR IMPLICATIONS ON SOIL CHARACTERISTICS USING REMOTE SENSING

By Mohamed El-Desoky Hereher

A Dissertation Submitted to the Faculty of the DEPARTMENT OF SOIL, WATER AND ENVIRONMENTAL SCIENCE

In Partial Fulfillment of the Requirements For the Degree of DOCTOR OF PHILOSOPHY In the Graduate College THE UNIVERSITY OF ARIZONA

2006

UMI Number: 3235007

UMI Microform 3235007 Copyright 2006 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code.

ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, MI 48106-1346

2

THE UNIVERSITY OF ARIZONA GRADUATE COLLEGE As members of the Dissertation Committee, we certify that we have read the dissertation prepared by Mohamed El-Desoky Hereher entitled MONITORING SPATIAL AND TEMPORAL CHANGES OF AGRICULTURAL LANDS IN THE NILE DELTA AND THEIR IMPLICATIONS ON SOIL CHARACTERISTICS USING REMOTE SENSING

and recommend that it be accepted as fulfilling the dissertation requirement for the Degree of Doctor of Philosophy. _______________________________________________________________________

Date: September 6th, 2006

Dr. Edward Glenn _______________________________________________________________________

Date: September 6th, 2006

Dr. Stuart Marsh _______________________________________________________________________

Date: September 6th, 2006

Dr. Alfredo Huete _______________________________________________________________________

Date: September 6th, 2006

Dr. David Hendricks

Final approval and acceptance of this dissertation is contingent upon the candidate’s submission of the final copies of the dissertation to the Graduate College. I hereby certify that I have read this dissertation prepared under my direction and recommend that it be accepted as fulfilling the dissertation requirement.

________________________________________________ Date: September 6th, 2006 Dissertation Director: Dr. Edward Glenn

3

STATEMENT BY AUTHOR This dissertation has been submitted in partial fulfillment of requirements for an advanced degree at The University of Arizona and is deposited in the University Library to be made available to borrowers under the rules of the library. Brief quotations from this dissertation are allowable without special permission, provided that accurate acknowledgement of source is made. Requests for permission for extended quotation from or reproduction of this manuscript in whole or in part may be granted by the head of the major department or Dean of the Graduate College when in his or her judgment the proposed use of the material is in the interests of scholarship. In all other instances, however, permission must be obtained from the author.

SIGNED: ________________________ Mohamed El-Desoky Hereher

4

ACKNOWLEDGEMENTS

I would like to express my appreciation to my committee members for helping and advising. My deepest gratitude goes to Dr. Ed Glenn, the major advisor, for giving me the confidence, support and chance to work under his guidance and advising. Special thanks go to Dr. Stuart Marsh for helping, educating and allowing me to use the facilities in the Remote Sensing Center, Office of Arid Land Studies. I would like to acknowledge my other two committee members, Dr. Alfredo Huete and Dr. David Hendricks who supported me greatly through the entire Ph.D. in the SWES Department. My sincere gratitude goes to all the SEWS Department staff, especially Judith Ellwanger and Veronica Hirsch for assistance. Both Grant Casady and Choy Huang in the Office of Arid Land Studies deserve thanks and appreciation for helping and giving a hand during the processing of the satellite images. I would like to thank every one in the SWES Department who provided me with any help and assistance during my study at the University of Arizona. My deep gratitude goes to my family: my beloved mother, my brothers and sisters for their support. Special thanks go to my sister Faten for the great help. I also would like to extend my gratitude to my darling wife Rasha for support and patience, and to my beloved cute daughter Noran. My deep thanks go to the Ford Foundation for funding my study and residence in the USA through the fellowship they granted me. I would like also to thank the Egyptian Government and Mansoura University for allowing me this chance to get my Ph.D. from the University of Arizona, USA.

5

DEDICATION I dedicate this dissertation to the spirit of my late father who passed away before the completion of my Ph.D. He was a great father and devoted his life for his family. May Allah the Almighty bless him and shower him with mercy.

6

TABLE OF CONTENTS ABSTRACT…………………………………………………………………......

9

INTRODUCTION……………………………….……………………………...

11

I-PROBLEM STATEMENT………………………………………………..

13

PRESENT STUDY……………………………………………………………...

15

I- THE NILE DELTA …………………………………………………….....

15

II-DISSERTATION OVERVIEW………………………………………….

17

REFERENCES………………………………………………………………….

19

APPENDIX A: INVENTORY OF AREAL EXPANSION OF AGRICULTURAL LANDS OF THE NILE DELTA USING LANDSAT SATELLITE IMAGES…………………………………………………………

21

I- INTRODUCTION……..…………………………………………………..

21

II-LITERATURE REVIEW………………………………………………...

24

Agricultural Land Estimations……………………………………………... Remote Sensing in Agricultural Studies…………………………………....

24 25

III- MATERIALS AND METHODS…..…………………………………...

27

Satellite Images……………………………………………………………… Atmospheric Correction and Radiometric Normalization…………………. Geometric Rectification, Resampling and Mosaicking…………………..… Normalized Difference Vegetation Index (NDVI)……………………….… Estimation of the Area of Agricultural Lands……………………………... Change Detection of New Agricultural Area…………………………….…

27 27 28 29 29 30

IV-RESULTS…………………………………………………………………

32

V- DISCUSSION..……………………………………………………………

33

Areal Extent of Agricultural Lands……………………………………….... Agricultural Expansion versus Urban Encroachment…………………….. Change in Agricultural Land Extent and its Implications on Soil Characteristics………………………………………………………………. Accuracy of Change Detection……………………………………………... VI-CONCLUDING REMARK……...………………………………………

33 34

VII-REFERENCES………………………………………………………….

38

35 36 37

7

TABLE OF CONTENTS continued APPENDIX B: MONITORING SPATIAL AND TEMPORAL AGRICULTURAL LAND DYNAMICS IN THE NILE DELTA, 1984 – 2003 USING NOAA-AVHRR IMAGES………………………………………

53

I-INTRODUCTION………………………………………………………….

53

II-MATERIALS AND METHODS…………………………………………

57

Data Set…………………………………………………………………….... Image Pre-processing……………………………………………………….. Linear Mixture Analysis (LMA)………………………………………….… Spatial and Temporal NDVI Variability along the Nile Delta…………….. Change Detection of Temporal NDVI Variability…………………………. NDVI and Seasonal Agricultural Change in Some Selected Locations…... III-RESULTS AND DISCUSSION..………………………………………..

57 58 58 59 60 61 62

Linear Mixture Analysis……………………………………………………. Spatial and Temporal NDVI Variability along the Nile Delta…………….. Change Detection of NDVI Variability…………………………………….. NDVI and Seasonal Agricultural Change in Some Selected Locations…... IV-CONCLUSIONS…………………………………………………………

62 63 65 67 69

V-REFERENCES……………………………………………………………

70

APPENDIX C: A CHANGE DETECTION STUDY OF THE GREATER CAIRO AREA AND ITS VICINITY USING LANDSAT REMOTE SENSING: A CASE STUDY OF AGRICULTURAL LAND DWINDLING. I-INTRODUCTION…………………………………………………………

88 88

II-MATERIALS AND METHODS…………………………………………

92

Study Area………………………………………………………………….... Satellite Data………………………………………………………………… Atmospheric Correction and Radiometric Normalization…………………. Geometric Rectification and Resampling…………………………………... Image Classification and Masking…………………………………………. Accuracy Assessment……………………………………………………….. Change Detection………………………………………………………….... a) Post-Classification Change Detection……………………………... b) Image Differencing……………………………………………….... c) Principal Component Analysis (PCA)……………………………...

92 93 93 94 94 95 96 96 96 97

8

TABLE OF CONTENTS continued III-RESULTS AND DISCUSSION………………………………………… Spectral Signature Curves and the Normalized Difference Vegetation Index (NDVI).......…………………………………………………………... Image Classification and Masking……………………………………….… Accuracy Assessment…………………………………………………….…. a) Sample Size for Accuracy Assessment………………………….…. b) Accuracy Assessment of Land Cover Maps…………………….…. c) Allowable Errors (E)…………………………………………….….

98

Change Detection………………………………………………………....… a) Post-Classification Change Detection……………………………... b) NDVI Image Difference…………………………………………..... c) Principal Component Analysis…………………………………….. Comparison of Change Detection Approaches…………………………….. Driving Forces for Land Use/Cover Change……………………….............

98 98 99 99 99 100 101 101 103 104 105 106

IV-SUMMARY AND CONCLUSIONS……………………..………….......

107

V-REFERENCES…………………………………………………………….

109

APPENDIX D: SOIL SALINITY PROBLEMS IN THE NILE DELTA…... I- INTRODUCTION…………………………………………………………

132 132

II- THE EXTENT AND CAUSES OF SOIL SALINITY…………………

132

III- CONTROL MEASURES FOR SOIL SALINITY…………………….

134

IV-REFERENCES…………………………………………………………...

136

9

ABSTRACT Egypt witnesses an increasing population growth concomitant with limited water and agricultural land resources. The objectives of this study were to utilize remotely sensed data for the inventory of agricultural lands in the Nile Delta, monitoring spatial and temporal variations in agricultural lands and quantifying agricultural land losses due to urbanization. Inventory of agricultural lands was designed using two approaches: thresholding and linear mixture analysis. We utilized 12 images from the Landsat satellite: 4 from Multi-Spectral Scanner (1972), 4 from Thematic Mapper (1984) and 4 from Thematic Mapper (2003) covering the entire Nile Delta. In addition, a set of 480 NDVI images were obtained from the Advanced Very High Resolution Radiometer (AVHRR) sensor that cover the period 1984-2003. Landsat images were subjected to atmospheric, radiometric and geometric corrections as well as image mosaicking. Normalized Difference Vegetation Index (NDVI) was applied and thresholding for agricultural land cover revealed that the areal extent of agricultural lands was 3.68, 4.32 and 4.95 million acres (one acre = 0.96 Egyptian Feddan) in 1972, 1984 and 2003, respectively. Linear mixture analysis of the AVHRR-NDVI with the TM-NDVI images showed that agricultural lands approached 4.11 and 5.24 million acres in 1984 and 2003, respectively. Using multitemporal Principal Component Analysis (PCA) for the TM and AVHRR images proved that reclamation activities were mostly along the western margins of the Nile Delta. Spatio-temporal analysis showed that middle delta has the highest agricultural vigor compared with the margins. Agricultural land loss was estimated in some cities within the delta as well as in Greater Cairo area. We studied the

10

land cover classification and change in Greater Cairo area based on 5 Landsat images acquired in 1972, 1984, 1990, 1998 and 2003. Agricultural lands lost 28.43% (32,236 acres) between 1972 and 2003 with an annual loss of 1040 acres. Agricultural lands on the peripheries of Cairo and its satellite towns were the most vulnerable areas. Soil salinization was another limiting factor for land reclamation. The main conclusion confirms that remote sensing is an accurate, efficient and less expensive tool for the inventory and monitoring agricultural land change in Egypt.

11

INTRODUCTION Egypt lies in the arid belt of the north-eastern corner of Africa, between latitudes 22o to 32o N and longitudes 25o to 34o E with an area of about one million km2. Geomorphologically, Egypt is divided into four units: the Nile Valley and Delta; the Western Desert; the Eastern Desert; and the Sinai Peninsula. Egypt is exclusively reliant on the River Nile for its water needs. This river is considered the artery of life for the Egyptians. It is the primary source for drinking, irrigation, and industry. It is also important for internal navigation. The agricultural land of Egypt is distributed along the narrow flood plain of the Nile and its delta. In the past, and for many centuries, the annual Nile flooding provided Egypt with sediment that formed one of the most fertile lands in the world. According to Ball (1952), the thickness of the sediment column in the flood plain varies from 6.7 m in the south of Egypt to about 11.2 m in the Nile Delta to the north. Before constructing the Aswan High Dam in 1964, an annual 84 billion m3 of water and 133.5 million tons of sediments were transported by the Nile in Egypt. From these sediments, 124 million tons/year discharged to the Mediterranean Sea, and 9.5 million tons/year settled in the flood plain (Stanley and Warne, 1993). After the 1952 Revolution, a new agricultural system was adopted in Egypt known as the Agrarian Reform. According to Beaumont et al. (1988), this system was applied when 40% of agricultural areas were held by less than 1% of the owners and 72% of the owners held 13% of the total agricultural lands. The Agrarian Reform introduced many laws to reduce the area owned by individual owners to only about 200 acres in

12

1952, 100 acres in 1962 and then 50 acres in 1969. As a result of this system, the majority of cultivated land fragmented and transferred from the landlords to the peasants. Limitations to agricultural expansions include water shortage, bad soil conditions and high reclamation costs. The Egyptian share of the Nile water has been the same since 1959 (55.5 billion m3), however, the erection of the High Dam at Aswan in 1964 helped regulate the water supply for irrigation throughout the year. Irrigation consumes 44 billion m3/year of water (El-Gibali and Badawy 1987). Soil salinization either due to natural or human-induced influence constituted the major obstacle to any reclamation effort. Abdel Hamid et al., (1992), noted that 80% of the salt-affected soils are distributed in the Nile Delta around the northern lakes. Reclamation costs range from LE 3000 to LE 8000 (USD 900.9 to USD 2402.4 in 1992) per acre contingent on the grade of the land, remoteness, and infrastructure (Biswas, 1993). In Egypt, there are three cropping seasons: winter (shitwi), summer (seifi), and autumn (nili) (Beaumont et al. 1988). The winter season extends from Nov. to May and the major crops cultivated are Egyptian clover, wheat, beans, and vegetables. The summer season extends from May to October and the main crops grown are cotton, rice, maize, and vegetables. The autumn season, refers to the period after the Nile flood month (Aug.) and the main crops are maize and vegetables (Ward, 1993). The average field size ranges from 1-4 acres for the majority of cultivated lands. Few fields are greater than 10 acres and all cultivable lands are vegetated (Tucker et al. 1984). The majority of cultivated fields are irrigated by flooding through conveying Nile water from the river course through tens of main canals and hundreds of irrigation ditches.

13

One of the most successful applications of remote sensing is in the agronomic field. The availability of at least three decades of digital data in multiple wavebands of the spectrum (visible, near infrared and thermal bands) and large ground coverage makes remote sensing superior to field-based studies. The premise in using digital data in monitoring agricultural lands is based upon the unique interaction of vegetation biomass with solar electromagnetic radiation which differs from other land cover signatures, e.g. water and bare deserts. Vegetation indices are quantitative surrogates of the vegetation vigor (Campbell, 1987). They are mathematical models enhancing the vegetation signal for a given pixel. Most vegetation indices were calculated from the reflection in two bands of the spectrum: the visible red and near infrared reflection (Aronoff, 2005). These indices could be utilized effectively in land cover change, monitoring vegetation cover density and crop identification (Bannari et al. 1995). The very famous Normalized Difference Vegetation Index (NDVI) is the model used in this study. I-PROBLEM STATEMENT A country of explosive population growth, Egypt has the lowest arable land per capita of any African country (Biswas, 1993). Population increased from 33 million in 1970 to about 70 million in 2000. Agricultural land increase is not comparable to population growth and census data are meager and even confusing. Egypt is implementing many ambitious national plans to reclaim and cultivate numerous desert areas in order to uphold food production for the fast growing population of the country. This step is crucial in minimizing the gap between crop production and crop import. According to Dennis (2001), the Egyptian imports of wheat made it the third largest

14

importer after China and Russia. Wheat imports increased from 1.2 million ton in 1961 to 7.4 million ton in 1998. In addition, maize imports increased from 0.1 to 3.1 million ton in the same period. Areas of reclamation potential include the fringes of the Nile Delta, the northern sector of Sinai Peninsula and the southern part of the Western Desert. Expansion along the fringes of the delta is governed by limited water resources, topography and soil conditions. At the same time populated centers are creeping upon neighboring agricultural lands. Estimates suggest that about 20,000 – 100,000 acres are being lost annually to urbanization and brick making (Daniels 1983, World Bank, 1990). For these reasons, there should be accurate estimations of agricultural lands and rates of agricultural land augmentations and losses. This is of utmost importance for planners and decision makers for the sustainability of development in Egypt.

15

PRESENT STUDY I-THE NILE DELTA The Nile Delta is one of largest and most well-known deltas in the world. Herodotus, a Greek Historian (5th century B.C.), was the first to describe the Nile Delta from its morphological features stating its resemblance to the Greek letter with its apex near the present Cairo in the south and its base along the Mediterranean Sea to the north. In past historical times, there were seven branches of the River Nile terminating into the Mediterranean Sea (Montasir, 1937 & Abu Al-Izz, 1971). These branches were (from west to east): Canobic (the farthest west); Bolbbitic (the present Rosetta Branch); Sebennytic; Phatnitic (the present Damietta Branch); Mendesian; Tanitic; and Pelusiac (the farthest east). Five of these branches silted up and disappeared leaving only two branches; Rosetta at the west and Damietta at the east (Snih and Weissbrod, 1973). The surface area of the Nile Delta represents only about 3 % of Egypt; however it stands for the bread basket for all Egyptians. The delta extends for 170 km from south to north and its width is 220 km. It is flat and slopes very gently, where the general slope between Cairo in the delta apex and the Mediterranean Sea in the north is only 12 m in 170 km (Abu Al-Izz, 1971). The Nile Delta is bounded by the Western Desert on the west and Suez Canal and the Eastern Desert on the east. The Nile Delta is exclusively dependent on the Nile water for irrigation, since rainfall is very scant except for the coastal region.

16

The climate of the Nile Delta is generally Mediterranean with hot summers and mild winters. Air temperatures average 18-19°C during winter and 30-31°C during summer. The coldest month is January and the hottest is August. The mean annual temperature generally increases as we move farther from the coast. Precipitation is generally scarce except along the Mediterranean littoral region, where annual rainfall reaches 200 mm. The inland areas receive very little precipitation; Cairo receives rainfall of only 22 mm/year (Beaumont, 1993), occurring between October and May. The mean potential evapotranspiration ranges from 570 to 1140 mm/year in the northern part of the delta and exceeds 1140 mm/year in the southern part (Beaumont et al. 1988). According to Stanley and Warne (1998), the Nile Delta belongs to the hyperarid climate region. Wind blows mostly from the north and northwest during summer and mostly from south and southwest during winter. Based on the FAO Soil Classification Map for Egypt (Beaumont et al. 1988), the major soils of the Nile Delta region and its fringes include Fluvisols (the alluvial plain soils), Solonchaks (salt-affected soils), Gravelly Ermolithosols (desert pavement), and Dynamic Ergosols (shifting sand dunes). In terms of soil texture, the majority of the delta soils are of dark heavy texture (40-70% clays) of alluvial parent materials (El-Nahal et al., 1977). The major clay minerals of the Egyptian alluvial soils formed by the deposition from the River Nile flooding are montmorillonite, illite and kaolinite which were derived from the eruptive rocks in the Ethiopian highlands (Hamdi 1967).

17

II-DISSERTATION OVERVIEW In Appendix A, the objective was to estimate the cultivated lands in the Nile Delta in three different dates; 1972, 1984 and 2003. Our hypothesis is based upon the fact that agricultural land cover has a higher NDVI value compared with urban, desert and water NDVI values. Consequently, NDVI thresholding for green biomass could help estimate areal expansion of agricultural lands. As one single image covers only about one quarter of the delta, we had to create a mosaic image consisting of four images from each of the MSS and TM sensors. The utilized images were acquired in the same season (summer), were geometrically rectified and were corrected from any atmospheric interference. The NDVI images were used to estimate agricultural land cover in 1972, 1984 and 2003 and to assess the agricultural land loss on the peripheries of some main cities in the Nile Delta, namely Mansoura, El-Mahala, Tanta and Zagazig. Moreover, we applied the multitemporal PCA in order to detect the direction of major land reclamations beyond the Nile Delta between 1984 and 2003. In Appendix B, we used NDVI data from 480 satellite images acquired by the AVHRR sensor covering the Nile Delta between 1984 and 2003. These data were obtained on a bi-monthly basis and were originally atmospherically corrected and geometrically rectified at NASA labs. We hypothesized that vigorous vegetation reflects higher long-term NDVI values and soil salinity lowers the NDVI value of agricultural lands. We applied a linear mixture analysis to estimate the areal expansion of agricultural lands of the Nile Delta using AVHRR images from 1984 and 2003 with the aid of TM images acquired in the same months. In addition, we created a one-image file compiling

18

the 480 images and then we generated a mean, a minimum, a maximum and a standard deviation images for the 480- layer NDV image in order to pursue spatio-temporal changes of the NDVI between 1984 and 2003. We applied a change detection approach using NDVI image differencing and multitemporal PCA to locate new cultivated lands and to monitor NDVI change. Finally, we selected four pixels from the mean NDVI AVHRR image representing different soil salinity and distributed in middle, west, north and east of the delta to monitor long-term variability of the NDVI in relation to different soil salinity levels. In Appendix C our concern was to estimate the agricultural and urban land cover/use change of the Greater Cairo area as about 20% of the total population in Egypt live in Cairo, and urban growth is the major threat to agricultural lands. The hypothesis was that any sudden decrease in NDVI along an urban/agricultural interface denotes a conversion of agricultural to urban land cover. We used five Landsat images obtained from the MSS (1972) and TM sensors (1984, 1990, 1998 and 2003). Atmospheric correction and geometric rectification was applied. Unsupervised classification was adopted to yield five land cover maps and a change detection using an integrated approach of post-classification, NDVI image differencing and multitemporal PCA was used to detect the agricultural/urban change. Finally, in Appendix D we explored some literature focusing on the nature, extent and control measures of soil salinity in the Nile Delta as a problem facing agricultural expansion.

19

REFERENCES

Abdel Hamid, M. A., Shreshta, D., and Valenzuela, C. 1992. Delineating, mapping and monitoring of soil salinity in the Northern Nile Delta (Egypt) using Landsat data and a geographic information system. Egypt. J. Soil Sci. 32, No.3, pp. 463-481. Abu Al-Izz, M. S. 1971. Landforms of Egypt. The American University Press in Cairo. Aronoff, S. 2005. Remote sensing for GIS managers. ESRI Press, Ca. Ball, J. 1952. Contribution to the geography of Egypt. Gov. Press, Cairo, Egypt. Bannari, A., Morin, D., Bonn, F., and Huete, A. 1995. A review of vegetation indices. Remote Sensing Review, 13: 95-120. Beaumont, P. 1993. Climate and hydrology. In, Craig, G. M. (edt.): The agriculture of Egypt. Oxford University Press, pp. 17-38. Beaumont, P., Blake, G, and Wagstaff, J. 1988. The Middle East. David Fulton Publisher, London. 623 pp. Biswas, A. 1993. Land resources for sustainable agriculture development in Egypt. Ambio, 22 (8): 556-560. Campbell, J. B. 1987. Introduction to remote sensing. The Guilford Press, New York, USA. Daniels, C. 1983. Egypt in the 1980s: The challenge. Economist Intelligence Unit Special Report No 158, London, 127 pp. Dennis, W. 2001. The role of virtual water in efforts to achieve food security and other national goals, with an example from Egypt. Agricultural Water management, 49: 131-151. El-Gibali, A. A. and Badawy, A. Y. 1978. Estimation of irrigation needs of Egypt. Egypt. J. Soil Sci., 18 (2): 159-179. El-Nahal, M. A., Abd El-Aal, R. M. Abd El-Wahed A. A. and Raafat, I. 1977. Soil studies on the Nile Delta. Egyptian J. Soil Sci. 17, 1, pp. 55-65.

20

Hamdi, H. 1967. The mineralogy of the fine fraction of the alluvial soils of Egypt. J. Soil Sci. UAR, 7 (1): 15-21. Montasir, A. H. 1937. Ecology of Lake Manzala. Bull. Fac. Sci., Cairo University, No. 12, pp. 1-50. Sneh, A. and Weissbrodm T. 1973. Nile Delta: the defunct Pelusiac branch identified. Science, 180 (4081): 59-61. Stanley, D., and Warne, A. 1993. Nile Delta: Recent geological evolution and human impact. Science, 260 (5108): 628-634. Tucker, C. J, Gatlin, J. A., Schneider, S. R. 1984. Monitoring vegetation in the Nile Delta with NOAA-7 AVHRR imagery. Photogrammetric Engineering & Remote Sensing, 50 (1): 53-61. Ward, P. N. 1993. Systems of the agricultural production in the Nile Delta. In Craig, M.M. (Edt.): The agriculture of Egypt. Oxford University Press. World Bank 1990. Egypt: environmental issue. Draft Discussion Paper. World Bank, Washington, D.C., 45 P.

21

APPENDIX A INVENTORY OF AREAL EXPANSION OF AGRICULTURAL LANDS OF THE NILE DELTA USING LANDSAT SATELLITE IMAGES I- INTRODUCTION Irrigated agriculture appeared in Egypt thousands of years ago. There are, however, no sustainable census data or accurate statistics of cultivated lands. Inventory of agricultural lands has been carried out mostly by ground surveys, government reports and personal estimations. Many studies give different estimates with some contradiction. The last formal agricultural census was carried out in 1961 (Biswas, 1993). The construction of the Aswan High Dam in 1964 provided additional water to reclaim about 1.35 million acres in the Nile Valley and Delta (Beaumont et al. 1988). Between 1952 and 1975, there were many political conditions that hindered the reclamation of new desert lands. In 1978, a Green Revolution was planned by the government to reclaim 2.9 million acres by the year 2000 (Barth and Shata, 1987). Reclamation projects were mainly in three regions: the margins of the Nile Delta, northwestern Sinai, and the southern sector of the country. According to Beaumont et al. (1988), between 1960 and 1980, there were six major reclamation areas around the Nile Delta and Sinai. These areas include: 1- El-Tahrir Province in the western Nile Delta; 2Maryut region near Alexandria; 3- Nubariya region south of Alexandria; 4- northern Nile Delta to the south of Lake Burullus; 5-south of Lake Manzala; and 6- northwestern Sinai region. New agricultural projects east of the Nile Delta and northern Sinai are irrigation

22

by Nile water (50%) mixed with drainage water to secure water needs (Weshelns, 2002). In addition, Egypt started a national plan to reclaim 1.43 million hectares (3.3 million acres) in the southern part of the country known as the Southern Valley Development Project (SVDP) (Elarabawy and Tosswell, 1998). Water needs will be delivered from Lake Nasser behind the High Dam at Aswan. Although Landsat Thematic Mapper (TM) data have been extensively used in local agricultural and vegetation stress studies, the normalized difference vegetation index (NDVI) from the TM images has seldom been used for regional agricultural mapping due to their narrow regional coverage (185 km) compared with the other coarser resolution Advanced Very High Resolution Radiometer (AVHRR) images (2048-km). The main objective of this study is to estimate the areal changes of agricultural lands of the Nile Delta for three decades demonstrating that remote sensing is a reliable means of monitoring agricultural lands. We used NDVI from MSS and TM instead of AVHRR to estimate the agricultural coverage in the Nile Delta for the following reasons: 1) the Landsat project started in the early 1970s with archived data at moderate spatial resolution that are not available from AVHRR; 2) the spatial resolution of these images (57 m and 28.5 m for MSS and TM, respectively) is very suitable to distinguish thousands of isolated communities (two to three houses) scattered in the agricultural fields of the Nile Delta which would not be detected by the coarser resolution AVHRR images; and 3)- Landsat, especially TM images, have more spectral bands in the visible, NIR and thermal regions of the spectrum than AVHRR images, thus providing more information.

23

The concept of change detection using satellite images is primarily based on quantifying the amount of change in radiance values, over some threshold, occurring in the same pixel of two images acquired in two different dates. These images should be obtained by the same sensor and should have the same pixel ground spacing, spectral characteristics and should be acquired in the same climatic conditions. Among the most common change detection techniques is the multi-temporal Principal Component Analysis (PCA), which is a linear transformation of correlated raw variables into uncorrelated variables or principal components. The resultant principal components are arranged according to their variance. The first component represents the large amount of variance (Jensen 1995, Hirosawa et al. 1996). The multi-temporal principal component analysis is applied to two or more stacked images, yielding different components each of which provides different information. The areas of change could be identified in the second, third, or consequent components. However, the difficulty in delineating the change and interpreting its occurrence represents the challenge in the multitemporal PCA application (Jensen, 1995).

24

II-LITERATURE REVIEW Agricultural Land Estimations The survey carried out by the Napoleon-led French Expedition estimated the agricultural lands in Egypt at the end of the eighteenth century as about 1.9 million hectares (4.71 million acres) and 60% of this cultivated areas occurred in the Nile Delta (Stanhill, 1981). El-Togy (1974) mentioned that about 3.7 million acres lie in the Nile Delta and this figure represents 60% of the total area under cultivation in the country. Sadek (1993) estimated the total expansion of the cultivated land in west of the Nile Delta for 35 years (1953-1988) to total 620,055 acres compared with 409,150 acres to the east of the Nile Delta for 33 years (1953-1986). Biswas (1993) reported that the Egyptian Public Authority for Survey assessed the total cultivated area in 1989 to be 7.8 million acres. Said (1993) reported, based on data from the Egyptian Central Agency for Statistics, that the total cultivated area of Egypt was 5.93 and 6.35 million acres in 1975 and in 1986, respectively. He also noted that the Remote Sensing Center of the Egyptian Academy of Science and Technology estimated the area to total 6.34 million acres in 1979. Abu-Zeid (1993) reported that the agricultural area of Egypt was 6.25, 6.33 and 7.47 million acres in 1970, 1980 and 1990, respectively. Fahim et al. (1999) stated that the cultivated areas were 6.34, 7.5 and 7.76 million in 1976, 1990 and 1995, respectively. Shalaby, et al. (2004) used two NOAA-AVHRR and SPOT images to monitor Egypt’s agricultural area in 1992 and 2000. They assessed agricultural area to be 8.69 and 9.96 million acres in 1992 and 2000, respectively. These contradicting estimates demonstrate the need for the current study using the most reliable method to date, remote sensing.

25

Remote Sensing in Agricultural Studies Marsh et al. (1992) mapped vegetation dynamics in the African Sahel using NDVI from NOAA-AVHRR and SPOT/XS images. Fung and Siu (2000) applied the NDVI derived from SPOT/HRV as a means for monitoring environmental changes in Hong Kong, China. Ramirez et al. (2004) studied the relationships between the vegetation index values and precipitation-growth stage cycles during 1996-1997 in Mesa Central, Mexico. Waweru et al. (2004) studied the desertification in Kenya’s dry land using NDVI from NOAA-AVHRR and TM images. Weiss et al. (2004) used AVHRR images to monitor NDVI variations in central New Mexico, USA. Principal component analysis has been widely used in multi-temporal change detection studies. Richards (1984) applied PCA to highlight regions of localized change resulting from bushfires using Landsat MSS images in New South Wales. Fung and LeDrew (1987) applied a multitemporal PCA for land-cover change detection. Hirosawa et al. (1996) utilized the multitemporal PCA to characterize vegetation communities in the State of Arizona based on AVHRR data. In Egypt, Tucker et al. (1984) studied the NDVI variations of the summer crops of 1981 in the lower Nile Delta using NOAA-AVHRR images. Salem et al. (1995) mapped the land cover classes in an agro-ecosystem area near Alexandria in 1990 using some vegetation indices and a supervised classification approach. They reported that the environmental vegetation index (EVI) and normalized difference vegetation index (NDVI) provided best greenness levels. They added that urban sprawl could eradicate agricultural lands within 70 years in that studied area.

26

Lenney and Woodcock (1996) studied the effect of spatial resolution on monitoring the status of agricultural land west of the Nile Delta. They observed that the overall accuracy is high with using coarse resolution data (120 m), however, fine resolution data (30 m), such as those from the TM are more accurate in extracting information about individual field productivity. Lenney, et al., (1996) using NDVI values of 10 TM images (1984-1993), estimated the reduction of agricultural productivity in the western delta region, due to high soil salinity, alkalinity, and water-logging to be 3.74% compared with 0.4% lost to urban expansion. They also mentioned that during the same period, the reclaimed areas in the Western Desert of Egypt increased by 43%. Sultan et al. (1999) monitored the urbanization of the Nile Delta using Landsat MSS and TM for the years 1972, 1984 and 1990. They discovered that the supervised classification is not successful. They, however, applied a manual classification with masking. They found that urban areas increased by 3.6, 4.7, and 5.7% of the Nile Delta for those periods. Lawrence et al., (2002) estimated the urban areas of Egypt using satellite MSS and TM images and a night-time imaging system and they reported that the urban areas of the country occupied only 3.7% of the total area. They also mentioned that over 30% of soils suitable for agriculture are under urban land use. Dewidar (2004) monitored the land use/cover changes along the coastal area of the Nile Delta using two satellite TM images acquired in 1984 and 1997. He applied image supervised classification for both images and assessed the amount of change for that region. He concluded that urban areas had been doubled and the agricultural areas had increased due to reclamation.

27

III- MATERIALS AND METHODS Satellite Images To study the entire areal extent of agricultural lands in the Nile Delta (Fig. 1), a set of Landsat MSS and TM satellite images were obtained for three different dates: 1972, 1984 and 2003. To cover the entire Nile Delta requires four Landsat MSS and four TM images, comprising a total of twelve satellite images for the three dates in question (Table 1). The satellite images were obtained from the United States Geological Survey (USGS) and the Global Land Cover Facility (GLCF) of the University of Maryland (WWW document retrieved in June 2005, (http://glcf.umiacs.umd.edu/portal/geocover/)). All twelve images were free of clouds, of high quality and acquired at the peak growth season in the Nile Delta, i.e. late summer. This was to ensure that all cultivable lands were cultivated, all the cultivated crops are in their climax phase, and all the images were obtained in approximately the same season and solar illumination angles. The major crops grown during the time of acquisition of satellite images are rice, cotton, maize and horticulture crops, such as tomato, potato, and cucumber. Atmospheric Correction and Radiometric Normalization Atmospheric correction is a process by which the degradation of the image quality caused by the influence of atmospheric interferences (dust, haze, smoke, etc) is cured, while radiometric correction involves the process of normalizing for variations resulting from sensor system degradation, the illumination source, or the sensor viewing angle (Aronoff, 2005). Atmospheric correction aims to retrieve the surface reflectance of each object by removing the impact of atmospheric effects. Atmospheric and radiometric

28

corrections for TM images were carried out in one step using the COST Model which is based on Chavez (1996) dark-object subtraction method. The inputs to this model were sun-earth distances, sun elevation angle, minimum digital number values (DN) of each image (bands 1-5 and 7). The output image was in reflectance values ranging from 0 to 1. For MSS images, the dark object subtraction approach was applied to all bands (1-4). The output images are in DN units. All the image processing techniques were applied using ERDAS Imagine 8.7 Software. Geometric Rectification, Resampling and Mosaicking Geometric correction is a process of warping the image to fit a planimetric grid or map projection. This process is crucial in remote sensing and change detection. By geometric rectification a given pixel remains at the same geographic location in all the images utilized in the study. In this study, the four TM images of 1984, which were obtained from the Global Land Cover Facility (GLCF) were used as the master images from which all the other images, including the MSS image, were rectified through imageto-image registration to the Universal Transverse Mercator Projection (UTM / zone 36 WGS 84) using a first-order polynomial transform. For each image registration, at least 20 prominent well-distributed ground control points (GCP) were chosen and a nearest neighbor resampling method was applied. The root mean square error (RMSE) ranged from 0.3 - 0.5-pixel for each registered TM image and less than one pixel for each MSS image revealing a higher precision of rectification. All the MSS images were resampled to have a pixel size equal to the TM pixels (28.5 m).

29

For each date, the four images were mosaicked to form a single image. In this process a histogram matching and a stitching process were applied. Thus, three final images representing the study area were obtained for 1972, 1984 and 2003. After that, a subset scene covering the Nile Delta and its fringes was created for each date (Fig. 2). As each image is processed independently, there was no need for further image normalization. Normalized Difference Vegetation Index (NDVI) Vegetation indices are surrogates to the abundance and activity of the green vegetation (Fung and Siu, 2000). The NDVI was calculated as (NIR-Red)/(NIR+Red) (Rouse et al. 1974), where, NIR is the reflectance or brightness in the near infrared portion and Red is the reluctance or brightness in the red portion of the spectrum. Theoretically, NDVI takes values from -1.0 to +1.0. Early studies using Landsat MSS mentioned that NDVI correlates significantly with the amount of green leaf biomass (Tucker 1979). Generally, positive values indicate green, healthy vegetation and negative and near zero values represent non-vegetated land-cover such as water, urban areas, and deserts. We applied the NDVI algorithm to each of the three mosaics. The NDVI image mean, maximum and minimum values were identified. Estimation of the Area of Agricultural Lands In order to accurately discriminate agricultural from non-agricultural land-cover, we had to pick a threshold in each NDVI image. A careful visual inspection for many points, especially at the margins of urban centers and at the delta fringes, was performed to determine the agricultural threshold in each NDVI image. Once the threshold was

30

determined, all the pixels equal to and greater than the threshold were considered as agricultural land, whereas the remaining pixels were considered non-agricultural lands. Finally, a binary image (consisting of two classes, 0 and 1) was created. Agricultural areas took the value of one; recoded in green color, and the non-agricultural areas took the value of zero; recoded in black color. The total agricultural lands of the entire Nile Delta in 1972, 1984 and 2003 were then obtained. In order to evaluate agricultural land loss due to urban sprawl in some selected locations, we assessed urbanization, based on the diminishing of agricultural lands around four main cities surrounded completely by agricultural lands within the Nile Delta namely: Tanta, El-Mahala, Mansoura and Zagazig. We obtained three equal subset images within the three binary NDVI images of 1972, 1984 and 2003 for each city and assessed urban and agricultural land area change in 1972, 1984 and 2003. Change Detection of New Agricultural Area We applied change detection to the TM subscenes of 1984 and 2003 images, because they were acquired by the same sensor and have the same spectral, spatial and radiometric characteristics. The objective of applying change detection was to locate the major land reclamations on both sides of the Nile Delta. Also, we compared the estimates of agricultural lands obtained by the binary NDVI images with the areas obtained from the multitemporal PCA change. Principal component analysis is a common data transformation technique for the compression of information content of a number of bands into just two or three principal components (Jensen 1995). It is used primarily to reduce redundancy in information

31

content in each image caused by the correlation between the spectral bands (Lillesand and Keifer 2000). In the literature, it is mentioned that in multi-temporal principal component analysis images, the PC1 represents the majority of information present in all images (no change) and PC2 and PC3 and the subsequent components represent the areas of land cover change (Aronoff, 2005). In this study, PCA was applied to a twelve-band image consisting of the stacked 1984 and 2003 TM images together. The first six components, obtained from the transformation, were examined carefully in order to identify the component highlighting the change between the two dates. The areas of temporal change, including urban and agricultural lands were identified and assessed.

32

IV-RESULTS The mean, maximum and minimum NDVI values of each image, as well as the threshold value of agricultural lands are shown in Table (2). Figure (3) presents a scatter diagram of the mean NDVI versus the threshold of agricultural lands. The total agricultural lands of the Nile Delta in the three binary NDVI images are shown in Figures (4&5). Figure 6 and Table (3) show the change of the urban/agricultural lands in the cities of Tanta, El-Mahala, Mansoura and Zagazig. In Figure (7-A), we present the first six principal components of the twelve-band (1984-2003) image. The third component (PCA-3) shows the locations of major changes (newly reclaimed projects and new urban areas). We recoded these areas in black (Fig. 7-B). Figures 8&9 and Table 4 show the agricultural lands obtained from the binary NDVI image of 1984 and 2003 and the areas of change obtained from the multi-temporal PCA-3 for the east and west sides of the delta.

33

V- DISCUSSION Areal Extent of Agricultural Lands It is worth mentioning that the mean NDVI value of the three images increased from 1972 (-0.08) through 1984 (0.21) to 2003 (0.30). This was accompanied by an increase in the threshold value of the agricultural lands in each image. The correlation coefficient (R2) was very high (0.99). The increased mean NDVI values in these three images may be attributed to the increase in agricultural productivity of crops in the Nile Delta which is accompanied by increased vegetation vigor and green biomass. According to Beaumont et al. (1988), in Egypt, the crop yield for cotton increased from 0.89 ton/acre in 1970/1974 to 1.14 ton/acre in 1984 and for maize from 1.57 ton/acre in 1970/1974 to 1.98 ton/acre in 1984. The spatial extent of the agricultural land was estimated at 3.68, 4.32 and 4.95 million acres in 1972, 1984 and 2003, respectively. Between 1972 and 1984, the amount of newly reclaimed areas was 0.64 million acres, with an annual rate of 53,333 acre. This is compared with new 0.63 million acres between 1984 and 2003 with an annual rate of only 33,157 acre. The total new agricultural lands between 1972 and 2003 are estimated at 1.27 million acres with a mean annual rate of 40,967 acres. According to El-Togy (1974), the agricultural lands in the Nile Delta constitute 60% of the total cultivated lands. Moreover, Biswas (1993), based on government estimates, reported that agricultural lands of the Nile Delta, in addition to Cairo and Giza agricultural lands, approach 62% of the total agricultural lands in Egypt. Based on these assessments, the total agricultural lands in Egypt are estimated by this study at 6.03, 7.08

34

and 8.11 million acres in 1972, 1984 and 2003, respectively (Table 5). These figures agree with many government reports and reveal high precision estimations. Agricultural Expansion versus Urban Encroachment Although agricultural lands in the Nile Delta increased by 1.27 million acres into the northern, eastern and western desert fringes between 1972 and 2003, urban sprawl on old fertile lands was growing forward. In fact, this figure (1.27 million acres) is a net value of what was added by reclamation and what was lost to urbanism. In order to get an overview of the agricultural land loss in the delta, we estimated the urbanization rate in four cities in the middle delta: Tanta, El-Mahala, Mansoura and Zagazig using the binary NDVI images for 1972, 1984 and 2003 (Table 3). All these cities witnessed an increase of their urban corridors upon neighboring agricultural lands. This resulted from the continuous population growth within these cities. The maximum loss of agricultural land between 1972 and 2003 was in Mansoura City (14% of the area) and the minimum was in Tanta City (8%). El-Mahala and Zagazig Cities lost 13 and 10% of their neighboring agricultural lands in the same period, respectively. On the other hand, Urban expansion was maximal in Zagazig City (19%) and 17% in Mansoura City. El-Mahala City increased by 13% and Tanta City witnessed the minimal increase (10%). If we take into account the other tens of main cities, hundreds of satellite towns, and thousands of villages distributed along the rural Nile Delta, we can get a sense of how much agricultural loss to urbanization occurred during the study period. It seems easier for people to build upon old agricultural lands than move to new cities in desert. The

35

agricultural loss around Greater Cairo area, the capital of Egypt hosting millions of people, is studied in details in Appendix C. Change in Agricultural Land Extent and its Implications on Soil Characteristics It is well known that information concerning landforms and soils can be inferred from satellite image analysis (Carrasco 2000). The trend of agricultural expansion in the Nile Delta was observed to be mostly in the western fringes of the delta. However, many reclamation projects were established to the east of the Nile Delta. The least direction was observed to the northern delta. Between 1984 and 2003, a total of 266,797acres were reclaimed to the west of the delta compared with 109,553 acres to the east. Soil condition and water availability are the most controlling factors of agricultural expansion. Soils in the northeastern part of the Delta are mostly of marine-alluvium (El-Nahal 1977) with higher salinity problems that require higher costs for reclamation as well as more water for salt leaching. Soils to the east are covered by shifting sand dunes. Their physical and chemical properties are not superior from the agricultural point of view. On the other hand, soils in the western fringes of the delta are mostly of calcareous soils originating from the limestone rocks in the vast Western Desert which were leveled by aeolian action (Beaumont, et al. 1988). Water is delivered to that area through many main canals, in addition to the ground water reserves. The soil quality may therefore be the reason for increased reclamation in the western side of the Nile Delta.

36

Accuracy of Change Detection In order to compare the accuracy of the change detection estimations obtained from the PCA with the estimations from the binary NDVI images, we have to be aware that the multitemporal PCA identifies all kinds of change, such as urbanization on the desert or on agricultural lands, reclamation of desert areas or even leveling of land for reclamation. On the other hand, the thresholded NDVI image identifies only the areal coverage of agricultural lands. For this study, the areas reclaimed in the western fringes of the delta between 1984 and 2003 were estimated by the binary NDVI image differencing to be 266,797 acres, whereas over the same period the PCA-3 change area is estimated at 296,104 acres. As the PCA determines all kind of change, the difference in these two values may be a result of other non-agricultural or pre-reclamation activities, such as leveling of land. The situation is a little bit different in the eastern fringes of the Nile Delta. There, in addition to reclamation projects, many new urban and industrial communities along the Cairo-Ismailia Road were established, e.g. the 10th of Ramadan City (Fig. 9). The agricultural areas added between 1984 and 2003 NDVI binary difference image total 109,553 acres. On the other hand, the area obtained from the change detection by the 1984-2003 PCA image totals 209,497 acres, which reveals a contribution of nonagricultural land in this estimation. The interpretation for this difference is that newly established urban communities were detected in the PCA image.

37

VI-CONCLUDING REMARK This study concludes that remote sensing could provide an indispensable tool for regional coverage, time-series, and high accuracy estimation of agricultural lands, especially in arid regions. The spectral and spatial characteristics as well as the low cost of Thematic Mapper (TM) images make it suitable for multitemporal change detection of agro-urban interactions. Although, there are ambitious plans to turn new desert regions into cultivated land, the continual sprawl of urban areas is devouring large amounts of fertile agricultural lands.

38

VII-REFERENCES

Abu Zeid, M. 1993. Egypt’s water resources management and policies. In Faris, M. and Khan, M. (edts): Sustainable Agriculture of Egypt. Lynne Rienner Publishers, Inc, Boulders, Colorado, pp. 71-79. Aronoff, S. 2005. Remote sensing for GIS managers. ESRI Press, Ca. Barth, H. K. and Shata, A. A. 1987. Natural resources and problems of land reclamations in Egypt. Wiesbaden, Germany. Beaumont, P., Blake, G, and Wagstaff, J. 1988. The Middle East. David Fulton Publisher, London, 623 pp. Biswas, A. 1993. Land resources for sustainable agriculture development in Egypt. Ambio, 22 (8): 556-560. Carrasco, C. P., Kubo, S., and Madhavan, B. 2000. Application of spectral mixture analysis of terrain evaluation studies. Int. J. Remote Sensing, 21 (16): 3039-3055. Chavez, P. S. 1996. Image-based atmospheric correction – revised and improved. Photogrammetric Engineering and Remote Sensing, 62 (9): 1052-1036. Dewidar, Kh. M. 2004. Detection of land use/land cover changes for the northern part of the Nile Delta (Burullus region), Egypt. Int. J. Remote Sensing, 25 (20): 40794089. Elarabawy, M. and Tosswell, M. 1998. An appraisal of the Southern Valley Development Project in Egypt. Aqua, 47 (4): 167-175. El-Nahal, M. A., Abd El-Aal, R. M., Abdel Wahed, A. A., and Raafat, I. 1977. Soil studies on the Nile Delta. Egypt. J. Soil. Science, 17 (1): 55-65. El-Togy, H. 1974. Contemporary Egyptian agriculture. Ford Foundation, Beirut. Fahim, M. M., El-Khalil, K. I., Hawela, F., Zaki, H. K., El-Mowelhi, M. N., and Lenney, M. P. 1999. Identification of urban expansion onto agricultural lands using satellite remote sensing: Two case studies in Egypt. Geocarto International, 14 (1): 45-47.

39

Fung, T. and Siu, W. 2000. Environmental quality and its changes, an analysis using NDVI. Int. J. Remote Sensing, 21 (5): 1011-1024. Fung, T., and LeDrew, E. 1987. Application of principal component analysis to change detection. Photogrammetric Eng. Remote Sensing, 53 (12): 1649-1658. Hirosawa, Y., Marsh, S. and Kliman, D. 1996. Application of standardized principal component analysis to land-cover characterization using multitemporal AVHRR data. Remote Sensing of Environment, 58: 267-281. Jensen, R. J. 1995. Introductory digital image processing. Prentice Hall, New Jersey. Lawrence, W. T., Imhoff, M., Kerle, N., and Stutzer, D. 2002. Quantifying urban land use and impact on soils in Egypt using diurnal satellite imagery of the earth surface. Int. J. Remote Sensing, 23 (19): 3921-3937. Lenney, M. P. and Woodcock, C. E. 1996. The effect of spatial resolution on the ability to monitor the status of agricultural lands. Remote Sensing of the Environment, 61:210-220. Lenney, M. P., Woodcock, C. E. Collins, J. B., and Hamdi, H. 1996. The Status of agricultural lands in Egypt: The use of multi-temporal NDVI features derived from Landsat TM. Remote Sensing of the Environment, 56:8-20. Lillesand, M. T. and Kiefer, R. W. 2000. Remote sensing and image interpretation. John Wiley & Sons, Inc. New York. Marsh, S. Walsh, J., Lee, C., Beck, L. and Hutchinson, C. 1992. Comparison of multitemporal NOAA-AVHRR and SPOT/XS satellite data for mapping land cover dynamics in the West African Sahel. Int. J. Remote Sensing, 13: 2997-3016. Ramirez, R. G., Trujllo, T. R. and Rodriguez, G.G. 2004. Analysis of NOAA-AVHRR images for crop monitoring. Int. J. Remote Sensing, 25 (9): 1615-1627. Richards, J. A. 1984. Thematic mapping from multi-temporal image data using the principal component transformation. Remote sensing of Environment, 16: 35-46. Rouse, J. W., Has, R. H., Schell, J. A., and Deering, D. W. 1974. Monitoring vegetation systems in the Great Plains with ERTS. Proceedings, Third Earth Resources Technology Satellite-1 Symposium, Greenbelt: NASA SP-351, 3010-3017.

40

Sadek, Sh. A. 1993. Use of Landsat imagery for monitoring agricultural expansion of East and West Nile Delta, Egypt. Egypt. J. Soil Sci. 33, No.1, pp. 23-33. Said, R. 1993. The River Nile, geology, hydrology and utilization. Pergamon Press, New York, USA. Salem, B., El-Cibahy, A. and El-Raey, M. 1995. Detection of land cover classes in agroecosystems of northern Egypt by remote sensing. Int. J. Remote Sensing, 16 (14): 2581-2594. Shalaby, A., Abol Ghar, M. and Tateishi, R. 2004. Desertification impact assessment in Egypt using low resolution satellite data and GIS. Int. J. Environ. Studies, 61 (4): 375-383. Stanhill, G. 1981. The Egyptian agro-ecosystem at the end of the eighteenth century-an analysis based on the “Description De L’Egypt”. Agro-Ecosystem, 6: 305-314. Sultan, M., Fisk, M., Stein, T., Gamal, M., Hady, Y., El-Araby, H., Madani, A., Mehanee, S, and Becker, R. 1999. Monitoring the urbanization of the Nile Delta, Egypt. Ambio, 28: 628-631. Tucker, C. 1979. Red and photographic infrared linear combination for monitoring green vegetation. Remote Sensing of Environment, 8: 127-150. Tucker, C. J, Gatlin, J. A., Schneider, S. R. 1984. Monitoring vegetation in the Nile Delta with NOAA-7 AVHRR imagery. Photogrammetric Engineering & Remote Sensing, 50 (1): 53-61. Waweru, M. N., Jahiah, M., Laneve, G. 2004. Spatial change analysis using temporal remote sensing and ancillary data for desertification change detection. Remote Sensing for Environmental Monitoring, GIS Applications, and Geology III, 5239: 345-356. Weiss, J. L., Gutzler, D. S., Coonrod, A., and Dahm, C. N. 2004. Long term vegetation monitoring with NDVI in a diverse semi-arid setting, central New Mexico, USA. J. Arid Environment, 58: 249-272. Weshelns, D. 2002. Economic analysis of water allocation polices regarding Nile River water in Egypt. Agricultural Water Management, 52: 155-175.

41

Figure (1): The Nile Delta of Egypt.

42

Table (1): The Satellite Data used in this study. Sensor Acquisition Path/Row Spatial Date Resolution TM 24 Aug. 2003 176/38 28.5 TM 24 Aug. 2003 176/39 28.5 TM 16 Sept. 2003 177/38 28.5 TM 16 Sept. 2003 177/39 28.5 TM 20 Sept. 1984 176/38 28.5 TM 20 Sept. 1984 176/39 28.5 TM 11 Sept. 1984 177/38 28.5 TM 11 Sept. 1984 177/39 28.5 MSS 5 Oct. 1972 189/39 57.0 MSS 31 Aug. 1972 190/38 57.0 MSS 31 Aug. 1972 190/39 57.0 MSS 19 Sept. 1972 191/38 57.0

Sun Elevation Angle 57 57 52 53 51 51 53 53 46 56 56 51

Source USGS USGS USGS USGS GLCF GLCF GLCF GLCF USGS GLCF GLCF GLCF

43

Figure 2. The three mosaic images of the Nile Delta obtained from the Landsat MSS and TM images of 1972, 1984 and 2003.

44

Table (2): The Statistical data of the NDVI images of the Nile Delta.

Image Year 1972 1984 2003

Mean NDVI -0.08 0.21 0.30

Max. NDVI 0.64 0.86 0.90

Min. NDVI -0.88 -0.65 -1.00

Threshold NDVI 0.14 0.21 0.32

Threshold value of ag. area

0.35 0.3 0.25 y = 0.4775x + 0.1809

0.2

2

R = 0.9952

0.15 0.1 0.05 0 -0.1

0

0.1 0.2 Mean NDVI of the image

0.3

Figure 3: A scatter diagram showing the relationship between the mean NDVI value of the image and the threshold of agricultural land.

45

Figure 4: Binary NDVI images showing the distribution of the agricultural lands of the Nile Delta in 1972, 1984 and 2003.

46

Area (millio n acre)

6 5 4 3 2 1972

1984

2003

Year Figure 5: The area of agricultural lands in the Nile Delta in 1972, 1984 and 2003 as measured from the satellite data.

47

El-Mahala Tanta

1972

1984

Mansoura Zagazig

2003

Tanta

El-Mahala

Mansoura

Zagazig

Figure 6: Columns A, B, and C show the 1972, 1984 and 2003 images, and rows 1,2,3, and 4 show the cities of Tanta, El-Mahala, Mansoura, and Zagazig.

48

Table 3: The areas, in acres, of urban and agricultural lands in the cities of Tanta, El-Mahala, Mansoura, and Zagazig between 1972 and 2003. 1972 1984 2003 Ag. (72-03) Urban (72-03) City urban Ag. urban Ag. urban Ag. acre % acre % Tanta 10427 16069 11019 15177 11447 14757 -1311 -8 1020 10 El-Mahala 6566 7788 6976 7271 7434 6765 -1023 -13 867 13 Mansoura 10224 14607 11841 12691 11915 12617 -1990 -14 1690 17 Zagazig 9922 20317 10488 19453 11776 18235 -2082 -10 1855 19

49

A A

New agricultural lands PCA-1

PCA-1

PCA-2

PCA-3

PCA-3

PCA-4

PCA-5

PCA-6

B

New agricultural lands

PCA-3 Figure 7: Change detection of the 1984-2003 mosaics: A) PCA showing the six components, and B) PCA-3 after thresholding shows areas of change.

50

NDVI-1984

NDVI-2003

PCA-3_1984-2003

Figure 8: Comparison of agricultural areas in the Western Delta obtained from the binary NDVI of 1984 and 2003 and the areas of change obtained by thresholding PCA-3 of (1984-2003).

51

New Urban areas

NDVI-1984

NDVI-2003

PCA-3_1984-2003

Figure 9: Comparison of agricultural areas in Eastern Delta obtained from binary NDVI of 2003 and 2003 and the areas of change obtained by thresholding PCA-3 of (1984-2003).

52

Table (4): The difference in area of reclaimed lands in west and east delta obtained from NDVI images of 1984 and 2003 and multi-temporal PCA image. NDVI-1984 NDVI-2003 NDVI-difference 1984-2003-PCA West Nile Delta 40,658 307,455 266,797 296,104 East Nile Delta 80,180 189,733 109,553 209,497 Table (5): Comparison of the total agricultural lands of Egypt by the present study and previous studies.

Agricultural Land, million acres 6.03 6.16 6.34 7.08 7.80 7.47 8.11 9.96

year 1972 1974 1976 1984 1989 1990 2003 2000

Reference This Study El-Togy 1974 Fahim et al. 1999 This Study Biswas 1993 Abu-Zeid 1993 This Study Shalaby et al. 2004

53

APPENDIX B MONITORING SPATIAL AND TEMPORAL AGRICULTURAL LAND DYNAMICS IN THE NILE DELTA, 1984 – 2003 USING NOAA-AVHRR IMAGES I-INTRODUCTION With the launch of the first generation of the NOAA-AVHRR satellites, potential remote sensing imaging of high temporal resolution for monitoring terrestrial vegetation emerged (Tucker et al. 1984). The premise behind agricultural applications of satelliteacquired data is that the chlorophyll of the green leaf strongly absorbs red radiation (630690 nm), whereas leaf cellular structure strongly reflects the near infrared radiation (760900 nm) (Tucker, 1979). Recent studies have shown that the NDVI obtained from the NOAA-AVHRR satellite provides useful information on vegetation dynamics as well as land cover (Justice et al. 1985, Prince 1991, Goward and Prince 1995, and Fuller 1998). Tucker (1979) noted that NDVI correlates strongly with green leaf biomass. In addition, NDVI correlates with leaf area index (Wiegand et al. 1979), crop vigor (Sellers 1985), chlorophyll content (Chappelle et al. 1992), and vegetation cycles (Weiss et al. 2001). The NDVI was developed by Rouse et al. (1974). It is computed from the visible and the near infrared (NIR) reflectance. In AVHRR data, the red (0.55 to 0.68 μm) and NIR (0.73 to 1.1 μm) reflectance values for a given pixel are set in the following equation: NDVI = (Ch2- Ch1)/(Ch2 + Ch1), where Ch1 and Ch2 are the red and near infrared reflectance of the AVHRR. Theoretically, NDVI values range from -1 to +1.

54

Many factors, other than green biomass, affect the NDVI of a given pixel. Atmospheric aerosols and water vapor lead to biased NDVI values (Justice et al. 1991). Sun illumination and sensor angles as well as sensor system modify the true pixel weight (Holben, 1986). Soil background, influenced by the texture; color; wetness and organic matter, is a serious problem influencing the NDVI of the canopy (Huete 1985). Generally, the NDVI decreases with increasing soil brightness (Huete 1985). The effect of soil background upon NDVI becomes negligible when the green cover approaches 70% or more of the pixel area (Nicholson and Farrar, 1994). New vegetation indices, e.g. Soil Adjusted Vegetation Index (SAVI) (Huete 1988) and Enhanced Vegetation Index (EVI) (Huete et al. 2002) were developed to avoid soil background interferences but they do not have a time series of global coverage (Hermann et al. 2005). The linear mixture analysis (LMA) approach seeks to quantify the weight of different spectral signatures combined linearly from different pure surface component (soil, water, vegetation, etc) “endmembers” present in one pixel (Metternicht and Fermont 1998). This approach aims to obtain information about land cover at a sub-pixel level (Coca et al. 2004). On the other hand, unmixing involves assessing the relative proportions of each surface component present in the pixel (Metternicht and Fermont 1998). Reference endmembers could be estimated from laboratory or field measurements (Tompkins et al. 1997). Moreover, Adams et al. (1993) mentioned that endmember spectra could be derived directly from image data. Many land-cover classification studies were conducted using the linear mixture model (e.g. Adams et al. 1995; Carrasco et al. 2000; Elmore et al. 2000 and Lu et al. 2003).

55

Time series analysis of the NDVI can provide information on seasonal and annual vegetation dynamics of an area assuming that land cover and soil characteristics in that area are in the same conditions (Schmidt and Karnieli 2000). Moreover, many NDVI models have been developed to describe vegetation health and vigor. For example, the Coefficient of Variation (COV) (Weiss et al. 2001) describes the ratio of the standard deviation value of a NDVI data set to the mean value of this data set. Weiss et al. (2001) showed that higher COV values reflect increased vegetation dynamics over the time period of investigation. They concluded that the COV provided satisfactory results on desertified areas of some rangelands in Saudi Arabia. In addition, Cogan (2001) developed a new index describing vegetation conditions. This index, the Vegetation Condition Index (VCI), describes the vigor and health conditions of vegetation. He observed that this index worked very well in monitoring a drought episode in Kazakhstan in 1996. Although he incorporated climatic and thermal band data in his vegetation health index, the VCI alone could be sufficiently informative of the vegetation health. Temporal change detection of the earth’s resources reflects the interactions and interprets the relationships between the natural resources and human impacts (Lu et al. 2004). Diverse techniques for change detection of land resources have also been reviewed (Singh 1989). Among these techniques are image differencing (Ridd and Liu, 1998) as pixel-to-pixel subtraction and multitemporal principal component analysis PCA (Sunar, 1998). The majority of change detection studies were carried out using data from MSS (Chavez and MacKinnon, 1994), TM (Guild et al. 2004), and SPOT (Fung and Siu,

56

2000) sensors. AVHRR data from the very course spatial resolution sensor were, however, seldom used. This study makes an endeavor to implement this approach. Ever since the 1980s, archived AVHRR-NDVI data have been used extensively in global as well as regional vegetation studies, e.g. Tucker et al.1983; Tucker et al. 1984; Hielkema, et al. 1987; Running and Nemani, 1988; Gupta, 1991; Gupta, 1992; Gupta et al. 1993; Farrar et al. 1994; Nicholson and Farrar 1994; Achard, 1995; Cihlar et al. 1996; Ramsey et al. 1995; Gitelson et al. 1998; Walker and Mallawaarachchi 1998; Ricotta et al. 1999; Schmidt and Karnieli 2000; Weiss et al. 2001; Boken et al. 2004; Dilley et al. 2004; Ramirez et al. 2004; Shalaby et al. 2004; Weiss et al. 2004; Anyamba and Tucker 2005; Herrmann et al. 2005; and Goetz et al. 2006. We concluded in Appendix A that agricultural lands of the Nile Delta as estimated from MSS and TM data were 3.68, 4.32 and 4.95 million acres in 1972, 1984 and 2003, respectively. The expansion occurred mostly in the western fringes compared with the eastern region and was lowest to the north. We attribute that to soil characteristics, availability of irrigation water and geographic locations in relation to major non-fresh water bodies, such as the Mediterranean Sea, Suez Canal and northern lagoons. In this Appendix, our primary objectives are to estimate the agricultural land cover using linear mixture analysis between TM and AVHRR images; to monitor the spatio-temporal NDVI variability (1984-2003) throughout the entire Nile Delta using AVHRR-NDVI data; and to study the effect of soil salinity on the NDVI seasonal variability of selected pixels (64 km2 each) distributed along the fringes of the Nile Delta (west, east and north) as well as in the middle delta.

57

II-MATERIALS AND METHODS Data Set The NDVI data used in this study were acquired from the Global Land Cover Facility at the University of Maryland (WWW document retrieved in Feb. 2006, (http://glcf.umiacs.umd.edu/data/gimms/)). The NDVI data from the AVHRR sensor, onboard the NOAA polar orbiting satellite, were processed by the Global Inventory Modeling and Mapping Studies (GIMMS) group at NASA’s Goddard Space Flight Center (see Tucker et al. 1994 & 2005). GIMMS NDVI data were obtained as two biweekly composites; one representing the first two weeks of the month and the other representing the second two weeks, totaling twenty four bi-weekly NDVI composites per year. The bi-weekly NDVI compositing was applied on the basis of maximum value compositing (MVC) in order to minimize cloud interferences, haze and dust contamination (Holben 1986; Anyamba and Tucker et al. 2005). These data have the advantage of being single band images. GIMMS NDVI images were georeferenced and corrected for sensor degradation, illumination and view angle variations, and atmospheric interferences (Huang et al. 1998 and Pinzon et al. 2004). Consequently, these data sets have a higher level of precision and accuracy (Herrmann et al. 2005). We used the African Continent file which was georeferenced to the Albers Conical Equal Area Projection with a Spheroid and Datum of WGS 84. The spatial resolution of these data is 8 km (Fig. 1). NDVI images are originally rescaled between -10000 and +10000. A total of 480 images covering the period from Jan. 1984 to Dec. 2003 were utilized. This period covers the same time span of the previous investigation using TM images (Appendix A).

58

Image Pre-processing Since the NDVI data sets we obtained from the Global Land Cover Facility of the University of Maryland were fully atmospherically corrected, normalized and georeferenced by the GIMMS group, very little preprocessing was required. We imported all the raw NDVI images of the whole African Continent to the ERDAS Imagine 8.7 environment. We created one image file consisting of 480 layers in which each layer represents the 15-day composite image during the whole study period (20 years x 24 image/year). In order to check the validity and consistency of the whole time series data set, one pixel in the Sahara Desert was chosen to monitor the 20-year variations of the NDVI values. We observed that the NDVI stayed the same, with only ± 6% change. Then we created a subscene image covering the entire delta. All the image processing procedures were conducted using ERDAS Imagine 8.7 and ArcMap 9.1 Software. Linear Mixture Analysis (LMA) We applied the LMA in order to obtain the fractional occurrence of agricultural lands of the Nile Delta using AVHRR-NDVI images for 1984 and 2003. The inputs to this analysis were the two binary TM NDVI images of the 1984 and 2003 obtained in Appendix A and the two AVHRR-NDVI images of Sept. 1984 and Aug. 2003 (the same acquisition times of the TM images). The first step was attaining the NDVI values for 10 AVHRR-NDVI pixels (64 km2 each) fairly distributed along the entire delta and representing different ranges of NDVI values, and the corresponding NDVI for equal areas obtained from both the TM in 1984 and 2003. The second step was estimating the

59

agricultural land cover fraction in each of the 10 areas for each TM image. The third step was creating a regression relationship between the NDVI of the 10 AVHRR pixels versus the percentage of agricultural land cover obtained from each area of the TM image. The fourth and last step was to estimate the areal extent of agricultural lands in both 1984 and 2003 AVHRR-NDVI images. The results were compared with the estimation of agricultural lands by thresholding the TM-NDVI images (See Appendix A). Spatial and Temporal NDVI Variability along the Nile Delta We created a mean NDVI image from the raw 480-layer image so that each pixel in the mean NDVI image has a mean value of 480 readings. Then we produced a spatial NDVI variability map for the entire Nile Delta. In the same manner, we estimated a standard deviation image for the 480-layer image. Then we divided the standard deviation of each pixel by its mean value to obtain an index image called the Coefficient of Variation (COV) image (Weiss et al. 2001). This index reflects agricultural dynamics, so that higher values refer to viable and more dynamic agriculture. In order to estimate the vegetation conditions throughout the study area, we created new minimum and maximum images for the 480-layer image and then we applied the Vegetation Condition Index (Cogan, 2001). In this index, the VCI = (NDVI – NDVImin) / (NDVImax – NDVImin), where NDVI is the mean value of each pixel, NDVImin and NDVImax are the minimum and maximum NDVI for the 480 readings of each pixel, respectively. Finally, we estimated the mean NDVI value of the entire delta for each year individually. We have thus obtained a twenty mean NDVI images for the twenty years (1984-2003).

60

Change Detection of Temporal NDVI Variability In order to pursue temporal change detection in agricultural lands in terms of NDVI variability over time, we applied (a) an image difference approach and (b) a multitemporal principal component analysis. In image differencing, we applied the algorithm of pixel-to-pixel subtraction (Singh, 1989) between the very first (1984) and the very last (2003) mean NDVI images in order to locate areas with increasing agricultural activities in terms of NDVI augmentation. We applied a threshold of 10% of change. Then we stacked each two biweakly images to form a single mean monthly NDVI image for each month for both 1984 and 2003. Finally, we subtracted each mean NDVI image for each month in 1984 from the corresponding month in 2003. For example, we subtracted the mean NDVI image of Jan. of 1984 from the mean NDVI image of Jan. 2003, the mean NDVI image of Feb. 1984 from the mean NDVI image of Feb. 2003, and so forth until the mean NDVI image of Dec. 1984 from the mean NDVI image of Dec. 2003. In addition, we applied a multi-temporal Principal Component Analysis (PCA) for the 480-NDVI-layer image in order to track the long-term NDVI changes over the entire Nile Delta. PCA has been widely used in multi-temporal change detection (Aronoff, 2005). As we mentioned in Appendix A, the first PCA always provides the information available in all the data set, whereas the second and consequent components highlight areas of spatial and temporal change. We examined the first six PCA images in order to determine the component which provides the temporal change in NDVI and consequently agricultural land.

61

NDVI and Seasonal Agricultural Change in Some Selected Locations The objective of this part is to address the variation of NDVI in relation to soil salinity, to designate the peak growth season at selected locations in the Nile Delta and the magnitude of seasonal NDVI at each location. We selected four pixels (8x8 km) distributed in west, east, north and middle delta. We determined the mean NDVI for each pixel over the twenty years of study. We used a soil salinity survey published for the Nile Delta (El-Nahal et al. 1977) and a Soil Classification for the Nile Delta (Beaumont et al. 1988) as ancillary data for this analysis. We created long-term NDVI temporal curves for each location.

62

III-RESULTS AND DISCUSSION Linear Mixture Analysis The primary objective of applying the LMA is to obtain the areal extent of agricultural lands from very coarse spatial resolution AVHRR images. As the pixel (8km) value of the AVHRR-NDVI image is a combination of spectral signatures from all land cover units present, higher agricultural land cover increases the NDVI value (Fig. 2). From Figure (3), the NDVI of the selected AVHRR pixels from the 1984 and 2003 years correlates strongly with the mean NDVI values of the equivalent areas in both the TM images of 1984 and 2003. The correlation coefficient is greater for 2003 image (0.95) than for 1984 one (0.60). In addition, the AVHRR-NDVI for both the 1984 and 2003 images correlates significantly with the percent of agricultural land cover obtained from the corresponding 1984 and 2003 TM NDVI images (Fig. 4). The correlation coefficient is 0.80 for the 1984 image compared with 0.83 for the 2003 image. From the regression equations, the agricultural lands in the Nile Delta in 1984 are estimated from the AVHRR data at 4.11 x 106 acres compared with 4.3 x 106 acres from the TM images (Appendix A). The AVHRR underestimated the agricultural lands by 4.4%. On the other hand, the agricultural lands were estimated at 5.24 x 106 acres in 2003 from the AVHRR (Table 1) versus 4.95 x 106 acres from the 2003 TM image. In this year, AVHRR overestimated the agricultural lands by 6.3% over the TM image. From these estimates, the linear mixture analysis could estimate agricultural land cover very precisely. The results are very comparable to the estimations obtained in Appendix A using TM data.

63

Spatial and Temporal NDVI Variability along the Nile Delta From the mean NDVI map (1984-2003) of the Nile Delta (Fig. 5), it is clear that there is a kind of spatial zoning of mean NDVI along the entire delta area, i.e. decreasing the mean NDVI values from the delta toward its margins. The highest mean NDVI value (>0.5) was observed for the bulk of the agricultural lands. This region is surrounded by a lower NDVI zone (0.4-0.5) from all sides of the delta. Areas falling in this class occur mainly along the western fringes of the delta. Areas of the third NDVI class (0.3-0.4) are generally smaller in area than the second class. These regions are located on the western and eastern margins and are sited in newly cultivated lands. Going farther to the east and west of the delta, the mean NDVI values decrease gradually. The lowest mean NDVI values are 0.3-0.2 and 0.2-0.1 and they represent areas either under reclamation or of lower quality soils (salt-affected soils). The Coefficient of Variation (COV) map of the Nile Delta (Fig. 5) depicts conspicuously that the majority of delta agricultural lands fall in the middle class (0.2– 0.3), whereas the delta margins have lower COV values (0.1–0.2). Very few pixels on the west and northeast of the delta have high COV values. From this distribution, it can be said that the delta margins are generally less vigorous than the middle delta in terms of COV values. The peripheries of the Nile Delta, including the Mediterranean Sea and the Western and Eastern Deserts have very low COV values (